Stochastic Policies
Types for representing randomized policies:
StochasticPolicysamples actions from an arbitrary distribution.UniformRandomPolicysamples actions uniformly (seeRandomPolicyfor a similar use)CategoricalTabularPolicysamples actions from a categorical distribution with weights given by aValuePolicy.EpsGreedyPolicyuses epsilon-greedy action selection.
POMDPPolicies.StochasticPolicy — Type.StochasticPolicy{D, RNG <: AbstractRNG}
Represents a stochastic policy. Action are sampled from an arbitrary distribution.
Constructor:
`StochasticPolicy(distribution; rng=Random.GLOBAL_RNG)`Fields
distribution::Drng::RNGa random number generator
CategoricalTabularPolicyrepresents a stochastic policy sampling an action from a categorical distribution with weights given by a ValuePolicy
constructor:
CategoricalTabularPolicy(mdp::Union{POMDP,MDP}; rng=Random.GLOBAL_RNG)
Fields
stochastic::StochasticPolicyvalue::ValuePolicy
POMDPPolicies.EpsGreedyPolicy — Type.EpsGreedyPolicyrepresents an epsilon greedy policy, sampling a random action with a probability eps or sampling from a given stochastic policy otherwise.
constructor:
EpsGreedyPolicy(mdp::Union{MDP,POMDP}, eps::Float64; rng=Random.GLOBAL_RNG)